Part 3

Investigate

The agentic spine — how runs become auditable scientific evidence. This part is for scientists: it frames each run as a Study with a pass/fail bar, rolls studies up into gated Investigations, and computes verdicts you can trust rather than assert.

Workspaces & the Workbench

The directory that is the model, and the AI-free server that turns it into a git-backed research notebook.

Studies

One question, one emit-contract, and one pass/fail bar wrapped around a composite.

Analyses, visualizations & report cards

The machinery that turns a wall of numbers into figures, derived tables, and graded scorecards.

Investigations

A gated DAG of studies that accumulates into a defensible claim under one research question.

Rigor & the evidence engine

The deterministic pipeline that computes a verdict from a run and writes the whole chain back for audit.

Working with AI agents

How the /viva-* skills drive the Workbench over plain HTTP, and how the work splits between agent and human curator.

Suggested path

Start with Workspaces & the Workbench for where the work lives, then Studies for the unit of evidence and Investigations for how studies combine. Analyses and Rigor explain how verdicts are produced; Working with AI agents shows who drives the loop.